A new VLM-based red teaming method shows that language-conditioned robot policies fail on diverse, contextually grounded instructions even when they ace existing benchmarks.
Playfusion: Skill acquisition via diffusion from language-annotated play
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Embodied Red Teaming for Auditing Robotic Foundation Models
A new VLM-based red teaming method shows that language-conditioned robot policies fail on diverse, contextually grounded instructions even when they ace existing benchmarks.